TL;DR: 44.2% of ChatGPT citations come from the first 30% of a page, the peak citation zone is the 10% to 20% band right after the title block, and the bottom 10% of any page earns just 2.4% to 4.4% of citations. Front-loading means putting the direct answer to the page's target query inside that early zone, then using the rest of the page for depth. Kill the throat-clearing intro, lead with the answer, and never save the reveal for the conclusion, because AI never reads that far.
What does "front-load the answer" mean?
Front-loading means the direct answer to the page's target query appears in the first 30% of the page, ideally within the first 150 to 200 words after the title, before any context-setting or storytelling. The rest of the page then earns its length with evidence, examples, and depth. It is the page-level application of the content capsule technique: if the capsule is the quotable block under each heading, the front-loaded intro is the biggest capsule on the page.
This is the fourth of the five structure rules from our guide to page structure for AI search, and it is the one with the hardest numbers behind it.
What does the data say about where AI reads?
Kevin Indig's analysis of 1.2 million ChatGPT responses and 18,012 verified citations found a distribution he calls the ski ramp: 44.2% of all citations come from the first 30% of a page, and the likelihood drops sharply after that. Burying key definitions deep in the content cuts retrieval probability by a factor of 2.5 compared to the introduction.
His follow-up study across seven verticals sharpened the map. The true peak is not the very top of the page: the 10% to 20% band performed best in every industry, because the first 10% is usually navigation, headline, and intro filler that the model skips. The bottom 10% earned just 2.4% to 4.4% of citations across all verticals, and conclusions were largely ignored. The steepness varies by industry, with finance the most extreme at 43.7% of citations landing in the first 30%, but the shape holds everywhere.
The likely reason is training. Large language models learned to read on journalism and academic writing, both of which follow the bottom line up front structure, so the model expects the most important information at the top and interprets everything after through that frame.
How do you front-load a page?
Four moves, applied to the zone the data rewards:
- Answer inside the first 150 to 200 words. After the title block, state the core finding or answer plainly: the number, the comparison, the implication. A TL;DR box does this job structurally and lands exactly in the peak band.
- Cut the throat-clearing. Openers about how the world is changing, how important the topic is, or what the article will eventually cover push the answer below the ramp. Delete them and start where they end.
- Define terms at first mention. Definitions are among the most cited sentence types, and a definition in the intro is 2.5x more retrievable than the same definition buried deep.
- Demote the conclusion. The final 10% is nearly invisible to AI, so never put a new fact, the hero stat, or the actual answer there. Use the closer for the human job it still does well: the summary restated and the call to action.
Doesn't front-loading kill the hook?
No, it replaces a hook that was already failing. The suspense structure, where a page builds context and delays the payoff, was designed for time-on-page metrics, not for readers. Buyers skim, journalists have led with the answer for a century, and a reader who gets the answer in the first screen and stays anyway is worth more than one held hostage by a delayed reveal. The depth does not disappear; it moves below the answer, where the genuinely interested keep reading and the machine finds its supporting evidence.
The honest trade-off is stylistic. Narrative arcs, cold opens, and slow builds still belong in storytelling and brand films. On a page built to win a query, they cost citations.
Front-loaded page vs narrative-arc page
| Signal | Front-loaded page | Narrative-arc page |
|---|---|---|
| First 150 words | The answer, with numbers | Scene-setting and stakes |
| Peak 10-20% band | Hero stat and definition | Still warming up |
| Intro's job | Deliver the bottom line | Delay the bottom line |
| Conclusion's job | Restate and call to action | Reveal the answer at last |
| AI citability | Answer sits where 44% of citations happen | Answer sits where 2 to 4% happen |
| Skimming reader | Served in one screen | Bounces unanswered |
How we apply this at HBS
Every article we publish opens with a TL;DR box that answers the page's query before the first heading, including this one, and the 44.2% finding is the reason our GEO playbook made it mandatory. When we audit client content, answer depth is the first check: we find where the primary answer actually appears on the page, and anything below the 30% mark becomes the first restructure. The material almost always exists. It is just parked where the model never looks.
Move the answer, keep the depth
Front-loading is a restructure, not a rewrite. Take the answer your page already contains, move it into the first screen, define your terms on arrival, and let the conclusion sell instead of reveal. If your best answers are sitting in the bottom half of your pages, our Advanced SEO Solutions team can run the answer-depth audit across your site, restructure the pages that matter, and put your key claims where both readers and AI engines actually look. Get a free audit and find out where your answers are hiding.
